Nonparametric regression with multiple variables?

[Rin Kraft]

Hi!
I'm running into a problem proving a relationship between wealth (y), amount of water used (x1), and time taken to gather the water (x2). I think a nonparametric multiple regression has the biggest chance of showing how these factors interact (rather than just doing a bunch of correlations) but I'm not sure what commands to use to get a) a scatter plot with the regression line and b) to get the analysis of the regression.
Any help would be much appreciated!
Thanks!
Rin

[FernandoRios]

Hi Rin,
If you have access to stata 15, there is the command -npregress- that does fully multivariate non-parametric regression using kernel smoothing methods.
Fernando

[Rich Goldstein]

you might wanted to see the blogs - urls are in
https://www.statalist.org/forums/for...ss-in-stata-15
note that if you don't have Stata 15, you need, as the FAQ explains, to tell us that

[Rin Kraft]

Hi! Thank you! I didn't notice that -- sorry. I have Stata 14 but it also seems to have the nonparametric regression option in that statistics tab.

[Steve Samuels]

All the non-parametric regression commands in the Statistic Tab take only a single predictor.
(I don't consider the quantile regression commands, "non-parametric", as the regression equation is looks like an f ordinary least squares equation; only the target parameters differ: quantile rather than mean).


An excellent substitute in Stata 14 for the new npregress is the contributed command KRLS (Kernel-based Regularized Least Squares)

In Stata, type:

ssc des krls
net install krls
net get krls  // to get the sample data set.

There is also an accompanying article. with theory and examples.